An auxfirst field guide · 2026 edition · Free to read
Why enterprise agent pilots stall between demo and adoption — and the five-layer playbook for getting across.
The chasm: the gap between the sponsored pilot and pragmatist adoption. Visionaries buy revolution. Pragmatists buy references, the whole product, and accountability — different tests, and the second one is the only one that counts.
This guide is for the people who own the gap. The executive who sponsored an agent pilot that impressed everyone and changed nothing. The product leader with three proofs-of-concept and no production deployment. The transformation lead who can feel the organization admiring agents from a safe distance.
If your company has bought agents but has not become agentic, you are standing at a chasm. This guide is about crossing it.
It builds on our pillar playbook, Five Steps to Becoming an Agentic Organization, and on a thirty-year-old idea from technology marketing that turns out to describe the inside of your company with uncomfortable precision.
Every large organization now has one. A folder, a Slack channel, a line item — somewhere the pilots go.
The demo went well. The vendor was credible. The innovation team was genuinely excited. The agent could summarize, draft, retrieve, act. And then, somewhere between the demo and the daily workflow, the thing quietly stopped being used. Not cancelled — worse. Orphaned. Still technically live, still on the invoice, no longer in anyone's habits.
Here is the uncomfortable pattern beneath these stories: the agent almost never failed technically. Capable agents are commodity now. The model is not the bottleneck. What failed is the crossing — the passage from the small group of people who wanted the agent to succeed to the much larger group of people who were supposed to work with it.
That passage has a name.
In 1991, Geoffrey Moore published Crossing the Chasm, and it became the canonical account of why promising technologies die on the way to the mainstream. Moore's insight was that the technology adoption curve is not smooth. Between the early adopters — visionaries who buy a dream and tolerate rough edges — and the early majority — pragmatists who buy proven solutions from safe hands — there is a gap so wide that most products fall into it. Visionaries buy revolution. Pragmatists buy references. Nothing you did to win the first group prepares you for the second.
Moore was describing markets. But the same curve runs through the middle of your organization, and agentic transformation lives or dies on it.
Look at your own company through this lens:
They were running agents on their own accounts before anyone approved anything. They forgive every hallucination because they can see where this is going.
The executive who greenlit the pilot. The product leader who sees a competitive edge. They will champion an agent that is 70% right because they are buying the trajectory, not the tool.
The operations manager. The account lead. The claims handler. The analyst whose name is on the output. Not hostile to agents — professionally skeptical, because when the agent is confidently wrong, they are the ones explaining it to a client, a regulator, or a boss.
The chasm inside your organization is the gap between the sponsored pilot and pragmatist adoption. Your pilot succeeded with people who wanted it to succeed. It has not yet been accepted by people who need it to be safe to rely on. Those are different tests, and the second one is the only one that counts.
This reframing changes the diagnosis. When adoption stalls, the reflex is to blame the technology (upgrade the model), the users (mandate the tool), or the timing (wait for the market to mature). Moore's frame says all three are wrong. Adoption stalls because you are selling to pragmatists the way you sold to visionaries — leading with capability and novelty, when pragmatists buy on completeness, accountability, and proof.
Moore's pragmatists have three well-documented buying behaviors, and every one of them translates directly to the employee deciding whether to fold an agent into their daily work.
In a market, peer companies. Inside a company, colleagues. The single strongest driver of internal agent adoption is a trusted peer saying, "I use it every day, and here is how it handles it when it's unsure." No launch email competes with that sentence.
Visionaries will assemble the missing pieces themselves; pragmatists expect the complete solution — support, training, integration, a clear escalation path. An agent dropped into a workflow without designed handoffs, without a named owner, without a way to say "I'm not confident here" is a partial product. Pragmatists do not adopt partial products. They wait.
They want to know who is responsible when the thing misbehaves — not philosophically, but by name. "The vendor" is not an answer. "IT" is not an answer. An agent with no accountable owner is, to a pragmatist, an unmanaged risk wearing a product's clothes.
Read those three again and notice what they have in common: none of them is a model capability. All of them are properties of the system around the agent. Which is exactly why the organizations that cross the chasm are not the ones with the best agents. They are the ones with the most complete system.
At auxfirst we map that system as five layers, and we wrote the operating manual for them in the five-step playbook. Here is how the layers line up against the chasm:
| Layer | What it is | What it answers for the pragmatist |
|---|---|---|
| 01 · Vision | A shared definition of "agentic," a chosen ambition per domain, a declared trust posture | "Is this a strategy or a fad someone will abandon?" |
| 02 · Ownership | Named owners for the agent, its adoption, and its risks — a pod, not a committee | "Who do I go to when it's wrong?" |
| 03 · Processes | The real workflow mapped — exceptions included — with autonomy set per decision point | "Does this thing understand how the work actually happens?" |
| 04 · Technology | A stack chosen to serve the process, validated against edge cases before launch | "Has anyone stress-tested this, or am I the stress test?" |
| 05 · AUX | The continuous trust layer: patterns, measurement, recovery — running across everything | "Can I rely on it — and will it tell me when I shouldn't?" |
The generic product is the agent. The whole product is all five layers. Skip one and the pragmatist's question it answers goes unanswered — and unanswered questions, for pragmatists, default to no.
The rest of this guide walks the crossing itself: how each layer becomes a chasm-crossing move.
Moore's prescription for crossing is famously narrow: don't attack the whole mainstream at once. Pick a single beachhead segment, win it completely, and let the references you earn there carry you to the adjacent ones.
Inside the organization, the beachhead is a domain — one workflow, one team, one part of the business where you will make the agent genuinely indispensable before you go anywhere else.
This is the discipline of the Vision layer, and it is where most enterprise programs fail first. The instinct — especially with executive enthusiasm behind you — is to be "agentic everywhere": a platform decision, a company-wide mandate, a hundred use cases in a backlog. That is attacking the whole front at once, and it produces the classic result: thin deployment everywhere, deep adoption nowhere, and a pragmatist majority whose only exposure to the agent was a mediocre first impression.
One domain, won completely, gives you the thing no budget can buy: internal references. Which is why the beachhead is not a small ambition. It is the only route to the large one.
Moore describes the chasm crossing as a D-Day operation: a focused force, total commitment, no ambiguity about who is doing what. The internal equivalent is the Ownership layer — and it is the layer organizations most reliably skip, because it feels like bureaucracy when the demo is exciting.
An agent that acts inside your business is a product with a user base. It needs product-grade ownership:
The pragmatist's accountability question — who do I go to when it's wrong? — is answered here, with names. When the escalation path is real and visible, the agent stops being an unmanaged risk and becomes a managed tool. That shift, invisible on any dashboard, is one of the largest single movements across the chasm.
The fastest way to lose a beachhead is to point an agent at a process you've never actually mapped. Most organizations run on two workflows: the documented one, and the real one — the one that lives in people's heads, in the exceptions, in the handoffs nobody wrote down. Automate the documented version and the agent breaks the moment it meets the real one. And it breaks in front of the pragmatists, which is the one audience you cannot afford a bad first impression with.
The Processes layer is chasm-crossing work disguised as diligence:
Trust climbs the ladder; it does not start at the top. The right first rung for a pragmatist audience is almost always lower than the technology allows.
Here is the chasm logic of the delegation ladder: every rung the agent earns in front of its users is a reference event. "It asked before committing." "It flagged the edge case." "It escalated exactly when it should have." These are the sentences pragmatists repeat to each other, and they are manufactured — deliberately — by process design.
Technology is the fourth layer on purpose. The most expensive mistake in enterprise agent programs is buying the platform first and reverse-engineering the work to fit it — stack-first transformation, the parade-ground army that has never seen the terrain.
For the crossing, three technology decisions matter disproportionately:
And now the layer that decides everything, the one that runs across all the others: AUX, the trust layer.
Recall what pragmatists buy first: references. In a market, references take years to accumulate. Inside an organization they move at the speed of lunch conversations — in both directions. One colleague saying "it saved me an hour and flagged the thing I'd have missed" is worth more than the entire launch communications plan. One colleague saying "it confidently invented a number in the client deck" undoes a quarter of goodwill by Friday.
This is why a safe agent isn't the same as a trusted one. Safety is a property of the machine — what it cannot do. Trust is a property of the relationship between your people and the agent — and it is earned in use, not configured at launch. AUX is the discipline of earning it deliberately:
Watch what happens when this compounds. Your beachhead pragmatists climb the ladder. They become references. References pull the adjacent domain across faster than the first one — Moore called this the bowling alley: each pin you knock down helps knock down the next. The second domain crosses in half the time not because the technology improved, but because the references existed.
Capability gets you to the edge of the chasm; trust is the bridge.
In a market of commoditized AI, features are replaceable and relationships are not. Trust is the moat — and inside your organization, it is also the road.
You do not cross the chasm on a launch date. You notice, in retrospect, that you crossed. The signs:
And the mirrored diagnostic — where the crossing is currently load-bearing on hope:
Any "no" on the first list, or "yes" on the second, marks your next move — and maps to a specific layer of the five-step playbook.
We are auxfirst — an agentic experience design agency. We work on the layer where agentic transformations actually succeed or fail: the trust and adoption layer that turns capable technology into systems people rely on.
The five layers map to how we engage:
| Layer | Engagement |
|---|---|
| Vision | Executive Seminar — align leadership on what agentic actually means, and choose the beachhead |
| Ownership | Executive Seminar + Team Workshop — from mandate to working ownership on the ground |
| Processes | Team Workshop + Blueprint Sprint — from process reality to an implementation-ready design |
| Technology | Blueprint Sprint + Agent Validation Sprint — design the spec, then prove it before launch |
| AUX | Advisory Retainer — and consulting baked into every engagement, because trust is the product |
Wherever you are relative to the chasm — still choosing the beachhead, or holding one and eyeing the next pin — it usually starts with a conversation.
Why transformations stall at the chasm — the strategy of the crossing, from beachhead to bowling alley.
Start a conversation →The operating manual: the plays, the ownership map, the delegation ladder, and the 60-second diagnostic — layer by layer, turn by turn.
Read the playbook →"Crossing the Chasm" is a framework by Geoffrey A. Moore (Crossing the Chasm, HarperBusiness, 1991). This guide applies its concepts to internal enterprise adoption; all interpretations are our own.